fulltext

Transcription

fulltext
Period integrals and other
direct images of D-modules –
Ketil Tveiten
Period integrals and other
direct images of D-modules
Ketil Tveiten
c
Ketil
Tveiten, Stockholm 2015
ISBN 978-91-7649-182-9
Printed in Sweden by Holmbergs, Malmö, Stockholm 2015
Distributor: Department of Mathematics, Stockholm University
Abstract
This thesis consists of three papers, each touching on a different aspect of
the theory of rings of differential operators and D-modules. In particular,
an aim is to provide and make explicit good examples of D-module direct
images, which are all but absent in the existing literature.
The first paper makes explicit the fact that B-splines (a particular
class of piecewise polynomial functions) are solutions to D-module theoretic direct images of a class of D-modules constructed from polytopes.
These modules, and their direct images, inherit all the relevant combinatorial structure from the defining polytopes, and as such are extremely
well-behaved.
The second paper studies the ring of differential operator on a reduced monomial ring (aka. Stanley-Reisner ring), in arbitrary characteristic. The two-sided ideal structure of the ring of differential operators
is described in terms of the associated abstract simplicial complex, and
several quite different proofs are given.
The third paper computes the monodromy of the period integrals of
Laurent polynomials about the singular point at the origin. The monodromy is describable in terms of the Newton polytope of the Laurent
polynomial, in particular the combinatorial-algebraic operation of mutation plays an important role. Special attention is given to the class of
maximally mutable Laurent polynomials, as these are one side of the conjectured correspondance that classifies Fano manifolds via mirror symmetry.
Sammendrag
Avhandlingen består av tre artikler, som belyser forskjellige sider av teorien om ringer av diffoperatorer og D-moduler. Spesielt har det blitt
lagt vekt på å gi konkrete eksempler på direkte bilder av D-moduler, for
å fylle et hull i den eksisterende literaturen.
Såkalte B-splines, en spesiell type stykkevis polynomielle funksjoner, er løsninger til direkte bilder av en klasse D-moduler konstruert
fra polytoper. Den første artikkelen gir en eksplisitt beskrivelse av disse modulene og deres direkte bilder. Modulene bevarer all den relevante
kombinatoriske strukturen til de definerende polytopene, og de oppfører
seg dermed veldig pent.
Den andre artikkelen handler om ringen av diffoperatorer på en redusert monomring (også kjent som en Stanley-Reisner -ring), også i tilfellet der grunnkroppen har positiv karakteristikk. De tosidige idealene
i ringen av diffoperatorer kan beskrives ut fra det tilhørende abstrakte
simplisialkomplekset, og flere fundamentalt ulike bevis gis.
I den tredje artikkelen beregnes monodromien av periodeintegraler
av Laurentpolynom om det singulære punktet i origo. Monodromien kan
beskrives ut fra Newtonpolygonen til Laurentpolynomet; den kombinatorisk/algebraiske operasjonen kalt mutasjon spiller en viktig rolle. Spesielt
omhandles klassen av maksimalt muterbare Laurentpolynom, som er interessant fordi disse er den ene siden av den formodede korrespondansen
som klassifiserer Fanomangfoldigheter via speilsymmetri.
List of Papers
The following papers, referred to in the text by their Roman numerals,
are included in this thesis.
PAPER I: B-splines, polytopes and their characteristic Dmodules
Ketil Tveiten, Preprint, arXiv:1306.6864, to appear in
Communications in Algebra
PAPER II: Two-sided ideals in the ring of differential operators on a Stanley-Reisner ring
Ketil Tveiten, Preprint, arXiv:1407.1643.
PAPER III: Period integrals and mutation
Ketil Tveiten, Preprint, arXiv:1501.05095.
Contents
Abstract
v
List of Papers
ix
Acknowledgements
xiii
1 Introduction
1.1 Rings of differential operators and D-modules . . . . . .
1.2 Distributions, B-splines, and polyhedral cell complexes .
1.3 Face rings and simplicial complexes, and ideals in rings of
differential operators . . . . . . . . . . . . . . . . . . . .
1.4 Toric geometry, mirror symmetry, and mutations . . . .
References
15
. 16
. 20
. 23
. 26
xxxiii
Acknowledgements
First and foremost I must thank my advisor Rikard Bøgvad, for all the
usual reasons; for teaching me everything I know about D-modules, for
always having somewhere new to point me when I got stuck or needed a
challenge, and for being such a cheerful and supportive guy.
Next in line are the organizers and participants of the PRAGMATIC
Summer School held in Catania, Sicily in 2013, in particular Alessio
Corti and Al Kasprzyk of Imperial College London, and Alessandro,
Andrea F, Andrea P, Diletta, Liana and Mohammad. This school was
the genesis not only of the third paper in this thesis, but of a productive
and continuing collaboration, and of great friendship and fun times, in
Catania, London, Udine and Ferrara.
The bread and butter of my time in Stockholm has of course been the
people at and connected to the Department of Mathematics. My fellow
inmates in room 105 in the time I’ve been here — Jens Forsgård, Per
Alexandersson, Elin Ottergren, Christine Jost, Stefano Marseglia, Elin
Gawell and the sofa in the back — have been great company. The staff
and administration of the Department have been great, in particular I
want to thank Samuel Lundqvist for helping me find a place to live when
I moved here. The Matlag deserves a special mention, as do Jörgen and
Jaja, for livening up the place and always having interesting stuff to say.
Finally, I want to thank all of the above, the climbing gang, and — in
no particular order — Ivan & Ornella, Christian, Björn, Håkon & Inna,
Sarah, Daniel, Qimh, Pinar, Iara, Tony, Gaultier & Claire, Rune, Felix
and Christoph for being great and/or entertaining company.
Finally I must thank my non-academic friends, and my family, for all
the joy they bring and support they give. In particular I want to thank
Martin & Anja for all their help in my early days in Stockholm; and my
parents, for everything.
1. Introduction
This thesis consists of three papers, each touching on a different aspect of
the algebraic theory of differential operators, and each taking advantage
of some combinatorial properties of the objects under study.
The first paper was inspired by the work of De Concini and Procesi
in [14] and [13], where they apply D-module techniques to study certain
particular B-splines and hyperplane arrangements. Noting that a Bspline is “essentially” a D-module theoretic direct image, and that good
explicit examples of D-module direct images are few and far between,
spelling out the details here would plug a gap in the existing literature.
Thus, the paper gives an explicit account of a family of D-modules that
models polytopes and hyperplane arrangements, and whose direct images
correspond to B-splines.
In the past, much work has gone into determining conditions for when
a ring of differential operators is simple, that is, having no nontrivial twosided ideals (see e.g. [25]). When the ring of differential operators is not
simple, it is in general difficult to describe the two-sided ideals, and this
is not a very well-explored problem. The aim of Paper II is to make
some headway by examining what in some sense is the simplest nontrivial case, namely reduced monomial rings. These rings, also known as
Stanley-Reisner rings, are in duality to abstract simplicial complexes, so
it would be natural to hope for some description of the ideal structure
of the ring of differential operators in terms of the associated simplicial complex. This indeed works out: two-sided ideals in the ring of
differential operators are in one-to-one correspondence with certain subcomplexes. While interesting on its own, this fact is perhaps secondary
to the methods used (several proofs are given), which one might hope to
generalize to less well-behaved contexts.
The third paper is somewhat different in scope; it is a part of an
ongoing program to classify Fano manifolds via mirror symmetry (see
[1]), in particular the dimension 2 case. One says that a Fano manifold
is mirror dual to a Laurent polynomial if a certain differential operator
constructed via quantum orbifold cohomology is equal to the differential operator that annihilates a particular period integral of the Laurent
15
polynomial. It is known in dimension 2 that every Fano manifold is
dual to a Laurent polynomial, i.e. the assignment “polynomials 7→ manifolds” is surjective (up to deformation); Paper III is concerned with
gathering evidence for injectivity of this assignment, namely showing
that nonequivalent Laurent polynomials produce nonequivalent differential operators. The paper computes, given a polynomial, the monodromy
of the associated differential operator at the origin; this is almost enough
to distinguish inequivalent polynomials, and some conjectures are made
about how to find the remaining information.
1.1 Rings of differential operators and D-modules
Let us begin by giving a quick summary of the beasts we about to slay:
rings of differential operators; and D-modules, which are modules over
those rings. For the interested reader, further material and greater detail
can be found e.g. in [5], [6], [10], [15], [17], [18], [22], and [23]. For the remainder of this section, Pn will denote the polynomial ring C[x1 , . . . , xn ],
I will be an ideal in Pn , and R = Pn /I is the quotient ring. The commutator of two elements in a ring is denoted by the bracket [a, b] := ab − ba.
Think first of Pn ; we can differentiate polynomials by applying the
∂
partial differential operators ∂i := ∂x
, which act by ∂i (c) = 0 for c ∈ C,
i
(
1 i=j
∂i (xj ) =
0 j=
6 i,
and satisfy the Leibniz rule: ∂i (f ·g) = f ∂i (g)+∂i (f )g, for any f, g ∈ Pn .
This formula can be rewritten as
(∂i f − f ∂i )g = (
∂f
)g
∂xi
∂f
or in other words [∂i , f ] = ∂x
, thought of as operators acting on Pn .
i
Extending Pn with variables ∂i obeying these relations, we get the Weyl
algebra, which is the ring of differential operators acting on Pn .
Definition 1.1.1. The Weyl algebra is the ring Chx1 , . . . , xn , ∂1 , . . . , ∂n i,
where the variables commute except for the relation [∂i , xi ] = 1.
We will denote the Weyl algebra by W , or if it is important how many
variables we work with, we let e.g. X := Cn , Y := Cm and denote the
Weyl algebras in n, m variables by DX , DY respectively, emphasizing the
connection to the geometry of affine space. Taking this a step further, if
16
X is a smooth variety (or complex manifold), we have associated to the
sheaf of reglar functions OX a sheaf of rings of differential operators DX ,
with DX (U ) = D(OX (U )) for each affine open set U . In particular, for
U ' Cn , we have DX (U ) ' W . We will switch back and forth between
these notations as appropriate in what follows.
It is not immediately clear what a “differential operator” acting on
R = Pn /I should be, but there is a property of the Weyl algebra that
generalizes in the right way. The Weyl algebra has a natural filtration by
the order of the differential operator: ∂ik has order k, xj has order zero,
and this clearly respects multiplication. Now observe that the Leibniz
rule implies that if δ has order k and f ∈ Pn is any element, then [δ, f ]
has order k − 1. With this in mind, Grothendieck gave the following
definition:
Definition 1.1.2. Let k be an integral domain, let A be a k-algebra,
let Dk0 (A) := A, and let Dkr (A) := {φ ∈ Endk (A)|∀f ∈ A : [φ, f ] ∈
Dkr−1 (A)}. The ring of k-linear differential operators on A is the ring
Dk (A) =
[
Dkr (A).
r≥0
In particular, for any finitely generated k-algebra R, i.e. a quotient
R = Pn /I of a polynomial ring, it can be shown that
D(R) = {δ ∈ W |δ(I) ⊂ I}/IW,
(1.1.1)
and this description does not depend on the presentation R = Pn /I.
The theory of D-modules is principally the theory of the modules of
the Weyl algebra, so when we say “D-module” this is what we mean unless otherwise specified. The motivation originally sprung from the study
of linear differential operators: δ ∈ W is a linear differential operator and
f0 ∈ Pn a solution to the differential equation
δ(f ) = 0
then the W -module W · f0 is isomorphic to the module W/Ann(f0 ),
where Ann(f0 ) is the left ideal in W of operators that annihilate f0 . In
this sense, solving differential equations in Pn is equivalent to studying
left ideals in W 1 . It soon turned out that D-modules were not merely
The analytical situation is equivalent: replace Pn by O, the ring of holomorphic functions in n variables, and W by D(O) = Oh∂1 , . . . , ∂n i, and essentially
the same results are true (the proofs may be very different, however).
1
17
an algebraic tool to study analytical problems, but an interesting theory
in their own right.
What follows are some of the most important results of the theory.
The dimension of a DX -module is defined via the Hilbert polynomial in
the usual way: the Weyl algebra has a natural filtration by total degree,
i.e. xa ∂ b has degree |a| + |b|. This gives an associated graded ring and to
each DX -module M an associated gr(DX )-module gr(M ), which has a
well-defined Hilbert polynomial p(n) = dimk (gr(M )n ). The dimension
of M is the degree of the Hilbert polynomial.
Theorem 1.1.3 (Bernstein,[4]). Let M be a nonzero D-module, then
the dimension is constrained by n ≤ dim(M ) ≤ 2n.
The inequalities are sharp: dim(W ) = 2n and dim(Pn ) = n, and
there exist modules of every intermediate dimension. Those D-modules
with dim(M ) = n are particularly interesting, and are called holonomic
D-modules. In particular, the regular holonomic 1 D-modules are extremely well-behaved; these include “most” of the D-modules one meets
in applications, e.g. modules of the form W · f for f ∈ Pn . In particular
every holonomic D-module is cyclic, i.e. generated by a single generator.
The celebrated theorem of Kashiwara tells us that regular holonomic Dmodules respect the geometry of their support exactly (the analogous
statement for OX -modules is very far from being true).
Theorem 1.1.4 (Kashiwara’s Theorem,[18]). Let X, Y be smooth schemes,
and let i : Y ,→ X be a closed immersion. Then the direct image functor
i+ gives an equivalence of categories between the category of regular holonomic DY -modules and the category of regular holonomic DX -modules
supported on Y .
The keystone and central result of the theory of D-modules is the
Riemann-Hilbert correspondence, which links D-modules, topology and
representation theory.
Theorem 1.1.5 (Riemann-Hilbert correspondence). Let X be a complex manifold. The bounded derived category of regular holonomic DX modules is equivalent to the bounded derived category of perverse sheaves
on X.
The sketch proof is as follows: locally a DX -module has a vector
space of solutions, these glue together to form a local system, and the
1
The definition of regularity here is quite technical, so I will not bore you
with it.
18
gluing data of a local system is equivalent to a representation of π1 (X).
The full proof fills half a book and is quite technical (see e.g. [6, chap.
VI-VIII]). We need to talk about direct images, and so we will need one
of these technical devices.
Definition 1.1.6. Let M be a DX -module. The de Rham complex
DRX (M ) of M is the cochain complex
M → Ω1X ⊗ M → Ω2X ⊗ M → · · · → ΩnX ⊗ M
P
where the differential is given by d(ω ⊗ m) = dω ⊗ m + i dxi ∧ ω ⊗ ∂i m.
Observe that for M = OX , the de Rham complex DRX (OX ) is the
same as the usual algebraic de Rham complex, and so we may speak of de
Rham cohomology of D-modules. The de Rham complex plays a central
role in the theory, indeed it is the content of the Riemann-Hilbert correspondence, the general statement of which is that the functor DRX (−)
(from the bounded derived category of DX -modules to the bounded derived category of perverse sheaves on X) is an equivalence.
To introduce direct images, we restrict to the simplest cases: inclusions of and projections to subspaces. These are the only cases we will
need for explicit computations.
Definition 1.1.7. Let X ' Y × Z, and let i : Z ,→ X, π : X Y
be the canonical inclusions and projections, respectively. Let N be a
DZ -module, and let M be a DX -module. Then
1. the direct image i+ N is defined to be N ⊗ OY , and
2. the direct image π+ M is defined to be DRX/Y (M ), where DRX/Y
is the relative de Rham complex Ω•Z ⊗ M (with only differentials
in the ∂z ’s and not the ∂y ’s).
Note that the direct image is not a DY -module, but a chain complex
of DY -modules; in other words the direct image lives in the derived categories. While the functors f+ are well-defined as stated above, to satisfy
the properties one expects the direct image to satisfy (e.g. composition
(f ◦ g)+ = f+ ◦ g+ , preservation of coherence and holonomicity), some
assumption of properness is often necessary; either that f : X → Y is
proper, or at least that f |Supp(M ) is. See for instance Paper III, where
the relevant morphism is proper, or Paper I, where it is proper when
restricted to the support of the module in question.
Kashiwara’s theorem makes direct images of inclusions rather trivial,
so let us consider the projection π : X = Y × Z Y . For O-modules,
19
the usual direct image π∗ M corresponds to the forgetful functor from
OX -modules to OY -modules (any OX -module is an OY -module via the
canonical map OY → OX ), so the “functions” in the direct image on
some open set are the functions defined on the inverse image of that set.
For DX -modules on the other hand, the elements of the direct image are
more like the functions defined by integration over the fibers of the map;
i.e. if f (x) = f (y, z) is a solution to a DX -module, then
Z
f (y, z)dz
(1.1.2)
(π∗ f )(y) =
π −1 (y)
is a solution to the direct image module. I say “more like”, as this is
the kind of analytic construction the direct image is intended to model;
precisely how well this intuition works is one of the topics of Paper I.
1.2 Distributions, B-splines, and polyhedral cell complexes
A common technique in the study of differential equations is looking at
weak solutions, that is, instead of studying functions f (x) e.g. on the
real line, one studies integrals
Z ∞
f (x)φ(x)dx
−∞
where φ(x) is a test function, a smooth function
R ∞ with compact support.
The point is that as φ has compact support, −∞ f (x)∂x φ(x)dx = 0, and
so the integral will satisfy the same differential equations as f . Taking
this idea to its logical conclusion, one shifts the viewpoint from studying
functions to studying linear operators that act on the space of test functions, which lets us solve more differential equations than can be done
with only smooth functions.R Such operators are called distributions, and
via the embedding f (x) 7→ f (x) · −dx the set of distributions includes
the functions we initially cared about. One often denotes the action of
a distribution g on a test function φ(x) by a bracket hg|φi. The reader
who is interested in the analytic theory of distributions is referred to
[16], though here the algebraic aspects are of course the important ones.
Distributions have a natural D-module structure, given by the action
hp(x)∂ α · g|φ(x)i = hg|(−1)|α| ∂ α (p(x)φ(x))i,
here p(x) is some polynomial and ∂ α = ∂1α1 · · · ∂nαn . In fact, it is a
theorem of Kashiwara ([17]) that any regular holonomic D-module is
generated by a distribution.
20
Example 1.2.1. Aside from smooth functions, distributions also encompass piecewise smooth functions, e.g.
(
x x>0
x+ =
0 x<0
R∞
is the distribution x+ : φ(x) 7→ 0 xφ(x)dx; piecewise continous functions, e.g. the Heaviside function
(
1 x>0
H(x) =
0 x<0
R∞
is the distribution H : φ(x) 7→ 0 φ(x)dx; the Dirac delta operator,
defined by δ(f ) = f (0) is a distribution. In fact, ∂x (x+ ) = H and
∂x (H) = δ.
For numerical applications, an important class of functions are the Bsplines, which are piecewise polynomial functions given by projecting a
polytope in some high-dimensional space Rm to some lower-dimensional
space Rs , and taking the volume of the fibers over each point; see e.g.
[8], [12] and [14]. Suppose σ ⊂ Rm is such a polytope, and π : Rm → Rs
is the projections, then the associated B-spline defined on Rs is given by
π∗ δσ (x1 , . . . , xs ) = vol(π −1 (x1 , . . . , xs ) ∩ σ)
or written another way
Z
π∗ δσ (x1 , . . . , xs ) =
σ∩π −1 (x)
dxs+1 · · · dxm .
This resembles the “solutions” to a direct image D-module, as in 1.1.2,
so to answer the question of whether it actually is, Paper I begins by
studying the obvious candidate: the distribution given by the characteristic function of σ. Let us introduce some notation: the affine hull of σ
is denoted by Hσ , the facets of σ are σ1 , . . . , σr and the outward normal
unit vectors of the facets are denoted by n1 , . . . nr
Definition 1.2.2. Let σ be a polytope1 in Rm . The characteristic distribution of σ is the distribution δσ defined by
Z
δσ (φ) =
φdx
σ
where dx is the restriction to Hσ of the standard measure on Rm .
1
By polytope I mean a closed contractible semialgebraic set defined by linear
polynomials
21
From Stokes’ theorem we have the following relations, which we call
the standard relations.
Proposition 1.2.3 (Standard relations). (i) If dim(σ) > 0, then for
any directional derivative ∂v where v is a vector tangent to Hσ , we
have
X
∂v · δσ = −
hv|ni iδσi .
i
(ii) Let I(σ) denote the defining ideal of Hσ . For any p ∈ I(σ), we
have
p · δσ = 0.
As σ is a polyhedral body, Hσ is an affine space defined by m −
dim(σ) equations of degree 1, and the corresponding polynomials
generate I(σ).
In light of the standard relations, we see that to study δσ we must
also study the facets δσi , their facets, etc; indeed the whole cell complex
σ
b made up of the faces of σ. In fact, it is convenient to consider cell
complexes with polyhedral cells in general, formed by gluing together
polytopes σ by their facets.
For the remainder of this section, we let X, Y and Z denote Rm , Rs
and Rm−s respectively (here sS< m), and DX , DY the Weyl algebras in
m and s variables. Let K = σ be such a cell complex; we define the
characteristic module of K to be the module
MK := DX · {δσ |σ ⊂ K}
generated by the characteristic distributions of all the cells in K. It is not
hard to see from the standard relations that the de Rham cohomology
of MK is related to the homology of K.
Theorem 1.2.4. The de Rham complex DRX (MK ) of MK is quasiisomorphic to the closed-support homology1 chain complex C•c (K) of K.
The paper contains an algorithm to compute the DX -annihilator
ideal of δσ that works via the algebraic Laplace transform, and an explicit
calculation shows a desirable but not obvious result:
Theorem 1.2.5. The module Mσb is generated by δσ .
1
22
Homology with closed support is also known as Borel-Moore homology.
The corollary is that MK can be generated by those generators corresponding to (locally) maximally-dimensional cells.
The sequence of subcomplexes Ki made up of all the cells of dimension ≤ i induce a filtration F0 ⊂ F1 ⊂ · · · Fdim(K) = MK which we
call the skeleton filtration. The skeleton filtration is a very useful tool
with which we can construct a canonical presentation (with generators
and relations) of MK as a DX -module, prove that the standard relations
generate all the relations between the generators, that MK is regular
holonomic, and that the functor K 7→ MK also preserves the cell complex structure: it preserves the poset of subcomplexes and the gluing
data of how the cells are attached to eachother.
Recall from 1.1.7 that the direct image of a projection is given by
π+ M = DRX/Y (M ). We are primarily interested in the lowest-degree
0 M , as this is where “functions” live (the higherterm of this complex, π+
degree terms of the complex encode cohomological information). If we
write X = Y × Z as before, with coordinates yi , zj on each factor, we
can express this module as
X
0
π+
M ' M/
∂zj M.
j
Geometrically, this corresponds to “flattening” the cells in the “vertical”
direction. Taking this quotient induces standard relations for the direct
image module, which appear identical to the set of recurrence relations
for the B-spline π∗ δσ defined by σ, proven by De Boor and Höllig [12].
These recurrence relations are the defining relations of a DY -module: We
denote the DY -module generated by the B-splines defined by the faces
of σ by SK := DY · {π∗ δτ |τ ⊂ σ}. From this the main result of the paper
follows, here “1-elementarily equivalent” is a slightly technical but fairly
mild topological condition.
0 M S , that
Theorem 1.2.6. There is a canonical surjective map π+
K
K
is an isomorphism is K is 1-elementarily equivalent to a cell complex K 0
with connected fibers.
1.3 Face rings and simplicial complexes, and ideals in
rings of differential operators
In Section 1.1 we discussed the properties of the Weyl algebra and the
ring of differential operators on a smooth variety, which locally looks
like the Weyl algebra. If the variety is singular however, the ring D(R)
may behave quite differently from the Weyl algebra. Notice that the
23
Weyl algebra is a simple ring, that is it has no nontrivial two-sided
ideals. This is due to the Leibniz rule [∂i , xi ] = 1, which implies that
whatever elements you try to generate an ideal with, you can act with
commutators — the commutator [a, b] is of course in both the two-sided
ideals generated by a and b — until you get a unit. This may no longer
hold when R is not regular, as we will see.
Let R = Pn /I, then recall from 1.1.1 that
D(R) = {δ ∈ W |δ(I) ⊂ I}/IW.
The important part here is that ∂i might not preserve I, so there may be
some room for two-sided ideals to live. Take e.g. I = hx1 x2 i in C[x1 , x2 ];
here D(R) is generated by operators of the form xai ∂ib with a ≥ 1, so
the ∂i are not in D(R). By explicit computation we can show that there
are three nontrivial two-sided ideals in D(R): the ideals hx1 i, hx2 i and
hx1 , x2 i.
What ideals occur and how they are related can of course get very
complicated, so in Paper II the idea is to study the simplest nonregular case, namely reduced monomial rings. These rings are also known
as Stanley-Reisner rings and have powerful combinatorial structure encoded by abstract simplicial complexes.
Definition 1.3.1. Let K be an abstract simplicial complex on vertices
x1 , . . . , xm . Then the face ideal IK in the ring C[x1 , . . . , xm ] is given by
IK = hxi1 · · · xir |{xi1 , . . . , xir } is not a subset of Ki.
The ring RK := C[x1 , . . . , xm ]/IK is the face ring or Stanley-Reisner
ring of K.
Returning to our example I = hx1 x2 i, this is the face ideal of the
simplicial complex with two isolated vertices x1 and x2 , and we can
notice that the ideals in D(R) each correspond to a subcomplex. The
idea of Paper II is to see what information about the ideals of D(RK )
can be read off the simplicial complex K; it turns out the ideals are
determined by an important class of subcomplexes called the stars of
simplices.
Definition 1.3.2. Let σ ∈ K be a simplex. The star of σ is the subcomplex
st(σ) := {τ ∈ K|τ ∪ σ ∈ K}.
The assignment σ 7→ st(σ) is inclusion-reversing.
24
Example 1.3.3. Let K be the complex on vertices x, y, z with face ideal
hxzi. It is a union of two 1-simplices, {x, y} and {y, z}, joined at a vertex,
y. The subcomplexes that are stars are st(x) = st({x, y}) = {x, y},
st(z) = st({y, z}) = {y, z}, and st(∅) = st(y) = K.
Example 1.3.4. Let K be the n-simplex ∆n . For any simplex σ ∈ K
we have st(σ) = K. The face ideal of K is the zero ideal, and the face
ring is the polynomial ring C[x1 , . . . , xn ].
Q
For each σ ∈ K, we denote xσ := xi ∈σ xi . We define the support of
a monomial xa = xa11 · · · xann to be the nonvanishing locus of the set of xi
such that ai 6= 0, i.e. the set of variables that appear in the monomial. It
is not hard to see that the closure of the support of any monomial is the
star of the simplex consisting of those xi that appear in the monomial.
A result of Traves [26] gives an explicit description of which monomials xa ∂ b that appear in D(R); in Paper I it is shown that there is an
equivalent formulation in terms of stars.
Proposition 1.3.5. Suppose supp(xa ) = st(σ) and supp(xb ) = st(τ ).
Then xa ∂ b ∈ D(R) if and only if st(σ) ⊂ st(τ ).
Using this and some explicit computation it is possible to show that
hxσ i ⊂ hxτ i if and only if st(σ) ⊂ st(τ ), and a key result:
Theorem 1.3.6. Any two-sided ideal in D(RK ) is generated by monomials xσ with st(σ) 6= K, and the ideals hxσ i generate (by sums and
intersections) the lattice of ideals.
From this follows the main result of the paper:
Theorem 1.3.7. The lattice of two-sided ideals in D(RK ) is exactly the
lattice of subcomplexes of K generated (by sums and intersections) by the
stars of simplices in K.
While so far all this is valid in any characteristic, in characteristic
p there is another quite different proof, based on the Frobenius automorphism: let q = pr , and let the subring of q’th powers be denoted by
Rq . Then R is an Rq -module in a natural way, and it is a theorem of
Yekutieli [27] that
[
D(R) =
EndRq (R).
q
Here EndRq (R) are the
can write
Rq -linear
R'
endomorphisms of R. In our case, we
M
q
(Rst(σ)
)mσ
st(σ)⊂K
25
where the sum is over all subcomplexes that are stars of some simplex
(including the empty one), Rst(σ) is the face ring of st(σ) and mσ are
some multiplicities depending on q. From this, we get
[ M
q
q
mτ
HomRq ((Rst(σ)
)mσ , (Rst(τ
)
D(RK ) =
))
q st(σ),st(τ )
and by explicit computation prove that for each st(σ) ⊂ K the ideals
q
q
generated by id : (Rst(σ)
)mσ → (Rst(σ)
)mσ (for each q) are equal, and
that these ideals — I call them J(st(σ)) — are the only nontrivial ones.
Theorem 1.3.8. The ideals J(st(σ)) generate the lattice of two-sided
ideals in D(RK ).
In fact, by a support argument we can see that J(st(σ)) = hxσ i.
The paper also includes a discussion of D-stable ideals, namely those
ideals of R that are sub-D(R)-modules. In [26], Traves describes the
D-stable ideals of RK as those formed by sums and intersections from
the minimal primes, and we obtain a new proof of this by applying the
following observation.
Proposition 1.3.9. Let J ⊂ R be an ideal, then J is D-stable if and
only if it is the restriction of a two-sided ideal in D(R).
1.4 Toric geometry, mirror symmetry, and mutations
We must introduce some basics of toric geometry. Sufficient for the
purposes of Paper III is the two-dimensional case, and I will entirely
skip the usual talk about torus actions as that is not relevant here; for a
thorough introduction to toric geometry see e.g. [11]. For this section,
we consider everything (lattices, polygons, polynomials) up to GL2 (Z)change of variables.
Consider the lattice Z2 , and pick two primitive elements a, b, and let
σ := R+ ha, bi ⊂ R2 be the convex cone they span. The set of lattice
points contained in this cone, σ ∩ Z2 , form a semigroup Nσ , generated
by the lattice points contained in the triangle formed by a, b and the
origin (including those on the boundary). If {a, b} is a lattice basis, then
Nσ ' N2 , and the semigroup ring C[Nσ ] is isomorphic to the polynomial
ring C[x1 , x2 ]; otherwise C[Nσ ] is nonregular, having a cyclic quotient
singularity. The spectra Xσ of the rings C[Nσ ] are the archetypical
affine toric varieties. Now partition the plane R2 into cones of this form,
so that two cones σ1 , σ2 intersect only in the ray spanned by a common
26
generator.
Such an arrangement is called a fan, and we denote it by
S
Σ := σi . One often describes fans by the rays that span the cones, i.e.
a fan is equivalent to the list of vectors that are primitive generators for
the cones in the fan. The affine toric varieties Xσi can be glued together
to form a variety XΣ ; this is a projective toric variety. The gluing goes
as follows: if ρ is the common ray of σ1 and σ2 , then the open sets where
ρ 6= 0 in Xσ1 , Xσ2 are identified.
Example 1.4.1. The fan Σ with generators (1, 0), (0, 1) and (−1, −1)
gives XΣ ' P2 .
Most important to us is varieties constructed in this way from the
normal fan of a convex lattice polygon. Let P be a convex lattice polygon
with vertices at primitive lattice points, with the origin as a strict interior
point — such a polygon is called a Fano polygon 1 . The inward normal
vectors of the edges form a complete fan ΣP , and so give a projective
toric variety XP . This variety contains all the information to reconstruct
P , indeed there is a canonically defined divisor DP on XP given by
X
DP =
hi DPi
where the DPi are the divisors corresponding to the edges of Pi and hi
are the lattice heights of those edges. The crucial thing for us is that
the global sections of DP correspond to the Laurent polynomials with
Newton polygon equal to P (see [11, 4.3.3]).
P
Given a Fano polygon P and a Laurent polynomial f = m∈P am xm
with N ewt(f ) = P , the classical period of f is the integral
Z
1
πf (a, t) =
ω,
C 1 − tf (a, x)
where C is the cycle |x1 | = |xR2 | = ε, and ω is the invariant volume form
on (C∗ )2 normalized to give C ω = 1; we think of the coefficients ai as
parameters. The period πf (a, t) is a multivalued holomorphic function
in a punctured disk about the origin, and we denote by Lf ∈ Cht, ∇t i
(here ∇t = t∂t ) the minimal-order minimal-degree differential operator
that kills it.
It is possible to explicitly compute Lf by various kinds of expensive
computations (see e.g. [21]), but this is not particularly enlightening.
More useful for purposes of proving things is the description of Lf as
the direct image module of the DXP -module generated by f , under the
1
The associated variety XP will be a Fano manifold whenever P is Fano; a
manifold is Fano if its anticanonical bundle is ample.
27
birational map τ : XP 99K P1 given by t = f1 .1 This gives us via some
general theory and the Riemann-Hilbert correspondence a nice description of the solution sheaf Sol(Lf ): it is the local system on P1 with stalk
H1 (Xt , Z) at t ∈ P1 , where Xt is the fiber of τ at t. It is in general very
hard to find all the singular points of τ and to find a good description of
the fibers there, but at t = 0 we have that X0 ' DP , and from this it is
easy to find a model for the general fiber Xt . Once we have the general
fiber it is easy2 to find the monodromy matrix at t = 0, which contains
very much of the interesting information about Lf .
Apparently unrelated, given a Fano manifold X with cyclic quotient
singularities that admits a Q-Gorenstein degeneration (or qG-degeneration
for short) to a toric variety, one can construct via quantum orbifold cohomology a differential operator QX ∈ Cht, ∇t i (see [24]).
We say that X and f are mirror dual if Lf = QX . Both these
differential operators are invariant under certain deformations; Lf is unchanged by mutation of f (I will define this below), while QX is unchanged by qG-deformation. It is thus a natural question whether the
equivalence classes are in bijection, as this would reduce the problem of
classifying Fano varieties (up to qG-deformation) to the much simpler
problem of classifying Fano polytopes with certain Laurent polynomials
on them; this is the conjectured mirror symmetry classification of Fano
manifolds (see [1], [2], [7]). Some progress has been made on proving
this conjecture:
Theorem 1.4.2 ([1]). The assignment P 7→ XP is a surjective map
from the set of mutation-equivalence classes of Fano polygons to the set
of qG-deformation-equivalence classes of Fano surfaces.
Why does this result talk merely about the polygons and not the
Laurent polynomials on them? There is a class of maximally mutable
Laurent polynomials that in a precise sense capture all the relevant information about P , and it is these that are the interesting ones in this
context (see Paper III and [20]).
The simplest cases of the general conjecture are proven, see [1], [9],
[19], and [24]:
Theorem 1.4.3. The set of qG-deformation classes of Fano surfaces
that are smooth or have only cyclic quotient singularities of type 13 (1, 1)
(respectively) is in bijection with the set of mutation-equivalence classes
1
Strictly speaking one resolves singularities of XP and base points of f to
get a morphism, and works with that instead.
2
See Paper III to evaluate how big you think this understatement is.
28
•
•
•
•
·
·
•
•
·
•
•
•
•
•
Figure 1.1: A polygon with 8 T -cones and one R-cone (grey) of type
1
3 (1, 1).
of maximally mutable Laurent polynomials on Fano polygons that are
smooth or have only R-cones of type 13 (1, 1) (respectively).
Let P be a Fano polygon as before, and consider an edge E of lattice
height h and lattice width w. We can write w = hk + r for some k ≥ 0
and ≤ r < h, subsequently we can subdivide the cone spanned by E
over the origin into k cones of width h and a cone of width r. A cone of
width equal to the height is called a T -cone, and one with width less than
the height is called an R-cone. From these we get a mutation-invariant
measure of P called the singularity content, defined as the pair (k, B),
where k is the number of T -cones in P , and B is the set of R-cones (these
concepts and terminologies are from [3]).
Example 1.4.4. The polygon pictured in Figure 1.1 has 8 T -cones,
unshaded, and one R-cone of type 31 (1, 1), shaded grey. The singularity
content of the polygon is (8, { 31 (1, 1)}).
So, what is this mutation of which I speak? Suppose an edge E has
a T -cone of height h. Then the mutation of P with respect to that cone
is given by removing from each positive height 0 < l ≤ h a slice of length
l, and adding at each negative height −l a slice of length l, as illustrated
in Figure 1.2.
This amounts to removing the T -cone on E, and adding a T -cone on
the opposite side. In this way one sees that the number of T -cones is
preserved by mutation; the fact that mutation with respect to an R-cone
does not make sense as defined here shows that the set of R-cones too is
invariant. The concept of mutation originally appeared in [2].
On Laurent polynomials there is a corresponding operation, provided
the coefficients satisfy some extra conditions. The mutation of P above
can be described in terms of a map on the ambient lattice of the normal
fan, that induces a map φ on the ring of rational functions C(N2 ), given
29
•
•
•
•
·
·
•
•
·
•
•
•
7→
•
•
•
•
·
•
·
•
•
•
•
•
•
•
Figure 1.2: Mutation of a Fano polygon: A T -cone is contracted, and a
new T -cone is inserted on the opposite side.
in suitable coordinates by (x, y) 7→ (x(1 + y), y). This map does not
preserve Laurent polynomials, but we say that a Laurent polynomial
f is mutable with respect to the mutation P 7→ P 0 if the image φ(f )
is a Laurent polynomial. Furthermore, f is maximally mutable if it is
mutable with respect to any mutation of P .1
Stated explicitly, this means that the slice fr of monomials “at height
r” must be divisible by (1+x)r at each positive height r, so the mutation
can be described by
fr 7→ fr (1 + x)−r ,
i.e. at positive heights r, we divide away the factor (1 + x)r , and at
negative heights −r we multiply by (1 + x)r .
2
Example 1.4.5. Let f = yx + 2y 2 + xy 2 + y1 . The Newton polygon has
vertices (−1, 2), (1, 2) and (0, −1); the cone spanned by (0, 0), (−1, 2) and
2
(1, 2) is a T -cone. Observe that we can write f = yx (1+x)2 +0(1+x)+ y1 ,
so f admits a mutation corresponding to this cone. The mutated polygon
has vertices (−1, 2), (1, −1) and (0, −1), and the mutated polynomial is
2
φ(f ) = yx + y1 + xy .
It is well known that a generic Laurent polynomial gives a curve of
genus equal to the number of internal lattice points of P . Imposing the
mutability condition for a T -cone of height h is equivalent to the curve
defined by f having a multiple point of multiplicity h. This drops the
genus by 12 h(h−1), which is equal to the number of internal lattice points
in the T -cone. Adding all this together we get an important result:
Theorem 1.4.6. The genus of a maximally mutable Laurent polynomial
with N ewt(f ) = P is equal to the number of internal lattice points of P
not lying internal to a T -cone.
1
30
The correct definition in higher dimensions is complicated, see [20].
We call this number the mutable genus, gmut , of P ; it is clearly
mutation-invariant. The Cauchy-Kowalewski theorem now implies that
the order (in ∇t ) of Lf and the rank of Sol(Lf ) both are 2gmut .
To compute the monodromy of Lf , we deform the fiber X0 at zero to
get the general fiber Xt , and find the monodromy via a number of local
computations. These come in two kinds: near the intersection points between components of X0 , corresponding to the vertices of P , the special
fiber looks in local coordinates like {xm y n = 0} (here m, n are the multiplicities of the intersecting components); this deforms to the smooth
curve {xm y n = t}. The other kind is deforming the components corresponding to the edges of P with R-cones on them; over such components
the general fiber has positive genus, and so there are a number of ramification points to resolve. What remains is then to construct a model for
an automorphism of a surface of the right genus, with the required order
and number of fixpoints (and we can show that any such models are
equivalent), and some power (we can find which) of this automorphism
is the right one.
Paper III is a part of the effort to prove the simplest non-smooth
case of the mirror symmetry correspondence, namely the case of only
1
1
3 (1, 1)-singularities, i.e. a singularity content of (k, {n × 3 (1, 1)}).
Theorem 1.4.7. Let P be a Fano polygon with singularity content (k, {n×
1
3 (1, 1)}), and let Xt be defined using a generic maximally mutable Laurent polynomial f with N ewt(f ) = P . Then there is a basis of cycles
{α, β, a11 , a12 , . . . , an1 , an2 } in H1 (Xt , Z) such that in terms of this basis, the
monodromy automorphism ω of H1 (Xt , Z) is given by
P
• ω(α) = α + (k + 2n − 12)β − nj=1 aj2 ,
• ω(β) = β,
• ω(aj1 ) = aj2 for 1 ≤ j ≤ n, and
• ω(aj2 ) = β − aj1 − aj2 for 1 ≤ j ≤ n.
In fact the methods in Paper III describe the monodromy at zero
for every possible singularity content, up to some minor ambiguity for
R-cones of width w ≥ 2. This ambiguity does not affect most important
information:
Theorem 1.4.8. The monodromy at t = 0 of Lf determines and is
determined by the singularity content of N ewt(f ).
31
32
References
[1] Mohammad Akhtar, Tom Coates, Alessio Corti, Liana Heuberger, Alexander Kasprzyk, Alessandro Oneto, Andrea Petracci, Thomas Prince, and
Ketil Tveiten. Mirror symmetry and the classification of orbifold del Pezzo
surfaces. arXiv:1501.05334, 2015. 15, 28
[2] Mohammad Akhtar, Tom Coates, Sergey Galkin, and Alexander M.
Kasprzyk. Minkowski polynomials and mutations. SIGMA Symmetry Integrability Geom. Methods Appl., 8:Paper 094, 17, 2012. 28, 29
[3] Mohammad Akhtar and
arXiv:1401.5458v1, 2014. 29
Alexander
Kasprzyk.
Singularity
content.
[4] I. N. Bernšteı̆n. Analytic continuation of generalized functions with respect
to a parameter. Funkcional. Anal. i Priložen., 6(4):26–40, 1972. 18
[5] Jan-Erik Björk. Analytic D-Modules and Applications. Kluwer Academic Publishers,
1993. 16
[6] A. Borel, P.-P. Grivel, B. Kaup, A. Haefliger, B. Malgrange, and
F. Ehlers. Algebraic D-modules, 2 of Perspectives in Mathematics. Academic Press
Inc., Boston, MA, 1987. 16, 19
[7] Tom Coates, Alessio Corti, Sergey Galkin, Vasily Golyshev, and Alexander
Kasprzyk. Mirror Symmetry and Fano Manifolds. In European Congress of
Mathematics (Kraków, 2-7 July, 2012), pages 285–300, November 2013. 28
[8] Elaine Cohen, T. Lyche, and R. F. Riesenfeld. Cones and recurrence relations for simplex splines. Constr. Approx., 3(2):131–141, 1987. 21
[9] Alessio Corti and Liana Heuberger. del Pezzo surfaces with
In preparation, 2015. 28
1
(1, 1)
3
points.
[10] S. C. Coutinho. A primer of algebraic D-modules, 33 of London Mathematical Society
Student Texts. Cambridge University Press, Cambridge, 1995. 16
[11] David A. Cox, John B. Little, and Henry K. Schenck. Toric varieties, 124
of Graduate Studies in Mathematics. American Mathematical Society, Providence, RI,
2011. 26, 27
[12] Carl de Boor and Klaus Höllig. Recurrence relations for multivariate Bsplines. Proc. Amer. Math. Soc., 85(3):397–400, 1982. 21, 23
[13] Corrado De Concini and Claudio Procesi. The Algebra of the Box-Spline.
arXiv:math/0602019v1, 2006. 15
[14] Corrado De Concini and Claudio Procesi. Topics in hyperplane arrangements,
polytopes and box-splines. Universitext. Springer, New York, 2011. 15, 21
[15] Alexandru Dimca. Sheaves in topology. Universitext. Springer-Verlag, Berlin, 2004.
16
[16] Lars Hörmander. The analysis of linear partial differential operators. I, 256 of
Grundlehren der Mathematischen Wissenschaften [Fundamental Principles of Mathematical Sciences]. Springer-Verlag, Berlin, second edition, 1990. Distribution theory
and Fourier analysis. 20
[17] Masaki Kashiwara. Regular holonomic D-modules and distributions on complex manifolds. In Complex analytic singularities, 8 of Adv. Stud. Pure Math., pages
199–206. North-Holland, Amsterdam, 1987. 16, 20
[18] Masaki Kashiwara. Algebraic study of systems of partial differential equations. Mém. Soc. Math. France (N.S.), (63):xiv+72, 1995. 16, 18
[19] Alexander M. Kasprzyk, Benjamin Nill, and Thomas Prince. Minimality and
mutation-equivalence of polygons. arXiv:1501.05335, 2015. 28
[20] Alexander M. Kasprzyk and Ketil Tveiten. Maximally mutable Laurent
polynomials. In preparation, 2015. 28, 30
[21] Pierre Lairez. Computing periods of rational integrals. arXiv:1404.5069v2,
2014. 27
[22] Philippe Maisonobe and Claude Sabbah, editors. Éléments de la théorie des systèmes différentiels. Images directes et constructibilité, 46 of Travaux en Cours [Works
in Progress]. Hermann, Paris, 1993. Papers from the CIMPA Summer School held in
Nice, August and September 1990. 16
[23] Bernard Malgrange. De Rham complex and direct images of D-modules.
In Éléments de la théorie des systèmes différentiels. Images directes et constructibilité
(Nice, 1990), 46 of Travaux en Cours, pages 1–13. Hermann, Paris, 1993. 16
[24] Alessandro Oneto and Andrea Petracci. Quantum periods of del Pezzo
surfaces with 13 (1, 1) singularities. In preparation, 2015. 28
[25] Karen E. Smith and Michel Van den Bergh. Simplicity of rings of differential
operators in prime characteristic. Proc. London Math. Soc. (3), 75(1):32–62, 1997.
15
[26] William N. Traves. Differential operators on monomial rings. J. Pure Appl.
Algebra, 136(2):183–197, 1999. 25, 26
[27] Amnon Yekutieli. An explicit construction of the Grothendieck residue complex. Astérisque, (208):127, 1992. With an appendix by Pramathanath Sastry. 25